Research On Fault Diagnosis Of Wind Turbine Gearbox Bearing Based On Improved Cost-sensitive Support Vector Machine

With the large-capacity and high parameter wind turbines putting into commercial operation, the requirement for real-time, accuracy and validity of the sets of equipment fault diagnosis is higher and higher. And Fault diagnosis is an important means of ensuring safe and reliable operation of the equipment. Frequent changes in wind speed, huge impact, variable load operating as its running characteristics, results in multi-type and high frequency failure to wind turbines. However gearbox as one of the wind turbine important transmission parts, also is prone to breakdown and cause the longest failure of wind turbines. This paper analysis the characteristics and deficiency of the traditional wind turbine gearbox diagnosis methods, and the cost-sensitive learning which can solve the classification imbalance problem was first applied to the fault diagnosis of wind turbine gearbox to explore a new method for fault diagnosis of gearbox.Aiming at the problem which training speed of CSVM is too slow when sample data is large amount, a ICSVM was proposed. The algorithm use KKT conditions to effectively filter the incremental samples set, and some samples that are useless in next training period are eliminated to achieve the boundary support vector set. The effectiveness of ICSVM was verified on the UCI data set. The specific implementation process based on wind turbine gearbox bearing fault diagnosis was achieved. The test results show that the proposed method has a lowest average misclassification cost, higher fault recognition rate, and faster training speed that fit for online fault diagnosis of wind turbine.To the issue of LSSVM without cost-sensitive property, CLSSVM was built. The different parameters of misclassification cost were embedded into optimization of primal LSSVM. The CLSSVM algorithm which aims to minimize classification costs was derived in detail. Finally, it is applied to the UCI standard data sets and the wind turbine gearbox bearing fault diagnosis The results of test illustrate that the proposed method which overcome the issue of LSSVM without cost-sensitive property has a lowest average misclassification cost, and can be able to enhance the accuracy of fault class samples, and diagnosis speed that fit for online fault diagnosis of wind turbine.